Add sv.RoundBoxAnnotator which annotates bounding box with round edges

[xaristeidou]

Description

This annotator will draw a bounding box which has round edges based on roundness attribute which accepts float values < roundness <= 1.0.

By default, roundness attribute percentage value is calculated based on the smaller size dimension (width or height). For example if height < width, and roundness is 1.0, the height dimension will be rounded completely.

This PR implements issue #693. (Fix #693)

Execution time of annotation process (tested on Ryzen 7 5800H):

• Exec time for 35 objects: 0.0018765926361083984
• Mean exec time for each object: 5.361693246023995e-05

No new dependencies required for this change.

• New feature (non-breaking change which adds functionality)
• This change requires a documentation update

I tested annotation results on walking people demo video. I used YOLOv8 to get the detection results. Screenshot of demo, using roundness = 0.5, is presented below:

Any specific deployment considerations

• Add new annotator class in Documentation.
• Add image example link reference in class Description in annotators/core.py

Docs

Docs not updated. Need to be updated with new annotator.

[onuralpszr]

@xaristeidou hello, first of it looks amazing. 🚀 I am taking the review.

[onuralpszr]

Simple Test Collab Link and python code

Google Collab link here

#!pip install git+https://github.com/xaristeidou/supervision.git@develop
#!pip install requests tqdm ultralytics

import supervision as sv
from supervision.assets import download_assets, VideoAssets
from ultralytics import YOLO

download_assets(VideoAssets.PEOPLE_WALKING)
model = YOLO('yolov8x-seg.pt')

frames_generator = sv.get_video_frames_generator(source_path='people-walking.mp4')
frames_iterator = iter(frames_generator)
frame = next(frames_iterator)

result = model.track(frame)[0]
detections = sv.Detections.from_ultralytics(result)

round_box_annotator = sv.RoundBoxAnnotator()
annotated_frame = round_box_annotator.annotate(scene=frame.copy(), detections=detections)
sv.plot_image(image=annotated_frame, size=(12, 12))

[onuralpszr]

cc @SkalskiP I did checks and write docs parts and also made collab part for testing so can you also check and upload the image to server so it can show up as well.

[SkalskiP]

@onuralpszr, can we make the name a bit shorter? I was thinking about RoundBoxAnnotator.

[onuralpszr]

@onuralpszr, can we make the name a bit shorter? I was thinking about RoundBoxAnnotator.

Okay let me handle it but you also please handle image for me (upload for docs collab already ready)

[onuralpszr]

@SkalskiP rename done. And test code and collab also changed as well.

[onuralpszr]

Thanks to @capjamesg's help he uploaded the pictures for docs so I am merging. Thank you all !